Signature-based Perceptual Nearness: Application of Near Sets to Image Retrieval

This paper presents a signature-based approach to quantifying perceptual nearness of images. A signature is defined as a set of descriptors, where each descriptor consists of a real-valued feature vector associated with a digital image region (set of pixels) combined with a region-based weight. Tolerance near sets provide a formal framework for our application of near sets to image retrieval. The tolerance nearness measure tNM was created to demonstrate application of near set theory to the problem of image correspondence. A new form of tNM has been introduced in this work, which takes into account the region size. Our method is compared to two other well-known image similarity measures: earth movers distance (EMD) and integrated region matching (IRM).